-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_test.py
227 lines (156 loc) · 8.71 KB
/
run_test.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
import torch
import pickle
from torch.utils.data import Dataset, DataLoader
import os
from torchvision import transforms
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torchvision.models as models
import csv
import argparse
import cv2
import numpy as np
from code.dataset_generator import get_test_loader
from code.networks import get_efficient_net
from code.data_cleaner import DataCleaner
from code.solution_values import SolutionValues, merge_solution_values, get_fixed_weights, get_custom_weights
import multiprocessing
import multiprocessing as mp
import glob
from tqdm import tqdm
CPU_COUNT = mp.cpu_count()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
def evaluate_net(net, test_loader, dataset):
net.eval()
solution_values = SolutionValues()
for data in tqdm(test_loader):
inputs, indices = data
inputs = inputs.to(device)
outputs = net(inputs)
for batch_index in range(len(indices)):
dataset_index = indices[batch_index].item()
metadata = dataset.image_metadata[dataset_index]
image_name = metadata.image_name
splitter = image_name.rsplit('_', 1)
ID, num = splitter[0], int(splitter[1])
for label in range(18):
value = outputs[batch_index][label].item()
solution_values.add_value(ID, num, label, value)
return solution_values
def get_b4_ensemble(test_loader, dataset):
print('B4 ensemble - 4 checkpoints:')
net = get_efficient_net('efficientnet-b4', 'final_models/b4-e12.pt', device)
solution_values1 = evaluate_net(net, test_loader, dataset)
net = get_efficient_net('efficientnet-b4', 'final_models/b4-e13.pt', device)
solution_values2 = evaluate_net(net, test_loader, dataset)
net = get_efficient_net('efficientnet-b4', 'final_models/b4-e15.pt', device)
solution_values3 = evaluate_net(net, test_loader, dataset)
net = get_efficient_net('efficientnet-b4', 'final_models/b4-e16.pt', device)
solution_values4 = evaluate_net(net, test_loader, dataset)
solution_values_b4_ensemble = merge_solution_values([solution_values1, solution_values2, solution_values3, solution_values4], [0.15, 0.35, 0.35, 0.15])
return solution_values_b4_ensemble
def get_b0_resize_ensemble(test_loader, dataset):
print('B0 resize ensemble - 4 checkpoints:')
net = get_efficient_net('efficientnet-b0', 'final_models/b0-resize-e12.pt', device)
solution_values1 = evaluate_net(net, test_loader, dataset)
net = get_efficient_net('efficientnet-b0', 'final_models/b0-resize-e13.pt', device)
solution_values2 = evaluate_net(net, test_loader, dataset)
net = get_efficient_net('efficientnet-b0', 'final_models/b0-resize-e14.pt', device)
solution_values3 = evaluate_net(net, test_loader, dataset)
net = get_efficient_net('efficientnet-b0', 'final_models/b0-resize-e15.pt', device)
solution_values4 = evaluate_net(net, test_loader, dataset)
solution_values_resize_ensemble = merge_solution_values([solution_values1, solution_values2, solution_values3, solution_values4], [0.20, 0.30, 0.30, 0.20])
return solution_values_resize_ensemble
def get_b0_resize_and_pad_ensemble(test_loader, dataset):
print('B0 resize and pad ensemble - 2 checkpoints:')
net = get_efficient_net('efficientnet-b0', 'final_models/b0-resize-and-pad-e12.pt', device)
solution_values1 = evaluate_net(net, test_loader, dataset)
net = get_efficient_net('efficientnet-b0', 'final_models/b0-resize-and-pad-e13.pt', device)
solution_values2 = evaluate_net(net, test_loader, dataset)
solution_values_resize_and_pad_ensemble = merge_solution_values([solution_values1, solution_values2], [0.5, 0.5])
return solution_values_resize_and_pad_ensemble
def get_solo_b0(test_loader, dataset):
print('B0 solo:')
net = get_efficient_net('efficientnet-b0', 'final_models/b0-solo.pt', device)
solution_values_solo_b0 = evaluate_net(net, test_loader, dataset)
return solution_values_solo_b0
def get_best_b0(test_loader, dataset):
print('B0 best:')
net = get_efficient_net('efficientnet-b0', 'final_models/b0-resize-and-pad-e13.pt', device)
solution_values_best_b0 = evaluate_net(net, test_loader, dataset)
return solution_values_best_b0
def solve_data_cleaner(image_name, nuclei_masks_folder, cell_masks_folder, dataset_folder):
dc = DataCleaner(nuclei_masks_folder, cell_masks_folder, dataset_folder)
value = dc.clean(image_name)
return image_name, value
def run(opt):
if not os.path.exists(opt.output_folder):
os.makedirs(opt.output_folder)
_, dataset = get_test_loader(opt.images_folder, opt.batch_size, opt.workers)
image_names = [x.image_name for x in dataset.image_metadata]
image_values = []
pool = mp.Pool(CPU_COUNT)
for image_name in image_names:
pool.apply_async(solve_data_cleaner, args=(image_name, opt.nuclei_masks, opt.cell_masks, opt.images_folder,), callback=image_values.append)
pool.close()
pool.join()
test_loader, dataset = get_test_loader(opt.dataset_folder, opt.batch_size, opt.workers)
border_and_garbage_value = {}
for ID, vals in image_values:
for i in range(len(vals)):
border_and_garbage_value[ID + '_' + str(i)] = vals[i]
print('RUNNING ' + opt.models + ':')
if opt.models == 'b0':
solution_values = get_best_b0(test_loader, dataset)
elif opt.models == 'final_ensemble_1' or opt.models == 'final_ensemble_2':
if opt.models == 'final_ensemble_1':
solution_values_b4_ensemble = get_b4_ensemble(test_loader, dataset)
solution_values_resize_ensemble = get_b0_resize_ensemble(test_loader, dataset)
solution_values_resize_and_pad_ensemble = get_b0_resize_and_pad_ensemble(test_loader, dataset)
solution_values_solo_b0 = get_solo_b0(test_loader, dataset)
if opt.models == 'final_ensemble_1':
solution_values = merge_solution_values([solution_values_b4_ensemble, solution_values_resize_ensemble, solution_values_resize_and_pad_ensemble, solution_values_solo_b0], [0.30, 0.30, 0.30, 0.10])
else:
solution_values = merge_solution_values([solution_values_resize_ensemble, solution_values_resize_and_pad_ensemble, solution_values_solo_b0], [0.4, 0.4, 0.2])
else:
raise Exception('invalid models!')
if not opt.no_postprocessing:
if opt.models != 'final_ensemble_1':
cell_weights, image_weights = get_fixed_weights()
else:
cell_weights, image_weights = get_custom_weights()
solution_values.calculate_negatives()
solution_values.weight_border_and_garbage_images(border_and_garbage_value)
solution_values.weight_cells_per_image(cell_weights, image_weights, border_and_garbage_value)
output_filename = opt.output_folder + opt.output_filename + '.xlsx'
output_sheet = opt.models
if opt.no_postprocessing:
output_sheet += '_np'
solution_values.to_output_table(output_filename, output_sheet, opt.append, opt.short)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('-i', '--images_folder', help='images folder', action='store', required=True)
parser.add_argument('-d', '--dataset_folder', help='dataset folder', action='store', required=True)
parser.add_argument('-o', '--output_filename', help='output filename', action='store', required=True)
parser.add_argument('-f', '--output_folder', help='output folder', action='store', required=True)
parser.add_argument('-c', '--cell_masks', help='cell masks folder', action='store', required=True)
parser.add_argument('-n', '--nuclei_masks', help='nuclei masks folder', action='store', required=True)
parser.add_argument('-m', '--models', default='b0', help='b0, final_ensemble_1 or final_ensemble_2', action='store')
parser.add_argument('-b', '--batch_size', default=8, help='batch size', action='store', type=int)
parser.add_argument('-w', '--workers', default=16, help='number of workers', action='store', type=int)
parser.add_argument('--no_postprocessing', help='used for getting gramdcam scores for b0', action='store_true')
parser.add_argument('-a', '--append', help='append to output file', action='store_true')
parser.add_argument('-s', '--short', default=None, help='store x decimal places', action='store', type=int)
opt = parser.parse_known_args()[0]
if opt.images_folder[-1] != '/':
opt.images_folder += '/'
if opt.dataset_folder[-1] != '/':
opt.dataset_folder += '/'
if opt.output_folder[-1] != '/':
opt.output_folder += '/'
if opt.cell_masks[-1] != '/':
opt.cell_masks += '/'
if opt.nuclei_masks[-1] != '/':
opt.nuclei_masks += '/'
run(opt)